Preview. These maps are our reading of public sources and haven't been checked with the projects' maintainers yet.
Agent memory
Memory layers and long-term memory systems for LLM agents.
22 projects59% active50% build on a shared standardMapped 2026-10-25
What they agree on
- How is it mainly sustained? company with a paid product, tier, or ads (76% of 21)
- Who is it for? developers (77% of 22)
- Does it work with models from different LLM vendors? yes, model-agnostic (86% of 22)
What they argue about
Is it actively maintained?
How is the main thing licensed?
Can it run entirely on your own device or infrastructure, without the maker's servers?
Do you need an account to use it?
Which shared standard or format does it mainly build on?
In the default setup, who mainly decides what gets remembered?
How can the person the memories are about see and edit them?
By default, how do old or wrong memories mainly go away?
Where do memories live in the default setup the docs lead with?
What is the main storage model for long-term memory?
What shape does it mainly take (the one its docs lead with)?
Is there a first-party MCP server for it?
How is memory separated between users or tenants?
Does it publish scores on public memory benchmarks (LoCoMo, LongMemEval, BEAM, etc.)?
Projects
| Project | Status | What it is |
|---|---|---|
| A-MEM | Quiet | Research 'agentic memory' (NeurIPS 2025): Zettelkasten-style notes in ChromaDB that the LLM links and evolves as new memories arrive. |
| Basic Memory | Active | Local-first knowledge base for AI assistants: Markdown notes with wikilinks that both you and the AI read and write over MCP, plus optional… |
| ChatGPT Memory | Active | Built-in memory in the ChatGPT consumer app: saved memories plus referencing past chats, managed by users in settings. |
| Claude memory tool (Anthropic API) | Active | Anthropic-defined client-side tool: Claude issues view/create/str_replace/insert/delete/rename commands on a /memories directory that your… |
| Cognee | Active | Open-source AI memory platform: turns documents, code and conversations into a knowledge graph plus vectors; local by default, Cognee Cloud… |
| Graphiti | Active | Open-source temporal knowledge graph framework from Zep: builds bi-temporal entity/fact graphs from episodes, on Neo4j, FalkorDB, Neptune o… |
| Honcho | Active | Plastic Labs' memory library and server for stateful agents: stores messages, reasons in the background, and maintains per-peer representat… |
| LangMem | Maintenance | LangChain's memory toolkit: hot-path memory tools for agents and a background memory manager, on LangGraph's BaseStore. |
| Letta (formerly MemGPT) | Changed direction | Originated MemGPT self-editing memory. Now Letta Code: a stateful agent harness whose agents rewrite their own memory blocks, skills and pr… |
| LongMemory (formerly OpenMemory) | Active | Local-first, self-hosted TypeScript memory engine: immutable, temporal, governed memory in SQLite with graph recall, served over npm, CLI,… |
| MCP reference memory server | Active | Reference MCP server: a local knowledge graph of entities, relations and observations stored in a JSONL file, edited by the model through M… |
| Mem0 | Active | Memory layer for agents and apps: library, self-hosted server, and hosted Mem0 Platform. LLM extracts facts from conversations into a vecto… |
| Memary | Quiet | Open-source memory layer for autonomous agents emulating human memory: knowledge graph (Neo4j/FalkorDB) plus memory stream and entity knowl… |
| Memobase | Quiet | User-profile-based long-term memory for chat apps: LLM maintains a structured profile and event timeline per user in Postgres. |
| Memori | Active | SQL-native memory layer that wraps your LLM client and extracts facts, preferences and agent traces into structured memory, via Memori Clou… |
| MemoryOS (BAI-LAB) | Maintenance | Research memory operating system (EMNLP 2025 oral) with short-, mid- and long-term persona memory tiers, plus a pip package and MCP server. |
| MemOS (MemTensor) | Active | Memory OS for LLMs and agents: unified add/retrieve/edit/delete API over a graph-structured memory, offered as cloud API, self-hosted servi… |
| Motorhead | Quiet | Early Rust memory server from Metal: sliding window of messages with incremental summaries and Redis-backed long-term search. |
| Redis Agent Memory | Changed direction | Redis's two-tier agent memory (session memory with TTL, plus extracted long-term facts with embeddings), now a managed Redis Cloud service. |
| ReMe (formerly MemoryScope) | Changed direction | From Alibaba's AgentScope team: local-first, file-based memory kit that stores agent memory as Markdown with frontmatter and wikilinks, sha… |
| Supermemory | Active | Hosted memory and context API plus consumer app: extracts facts, builds user profiles, handles contradictions and forgetting; a local singl… |
| Zep | Active | Hosted agent memory / context engineering platform built on the Graphiti temporal knowledge graph, with its own graph database service (Kon… |
Where they meet
Not mapped yet.
Follow
RSS for this community: projects added, status changes, and later polls and discussions.
Follow the whole community: OPML of every project's own release feed (where we found one) · OPML of every project's Poll Position feed.
The poll
Every project above was answered against these questions. Core questions are shared by all communities; the rest are this community's own.
- Is it actively maintained? core
active · active, but changed direction (pivot, fork, or rename) · maintenance or slowing · dormant, sunsetting, or archived - How is the main thing licensed? core
permissive open source (MIT/Apache/BSD) · copyleft or source-available with restrictions · open parts around a closed core · proprietary or closed source - How is it mainly sustained? core
individual side project, no revenue · individual with revenue (paid app, tier, ads, or donations) · community, foundation, or research · company with a paid product, tier, or ads · company or startup with a free product (no revenue yet) · big company side project - Can it run entirely on your own device or infrastructure, without the maker's servers? core
yes · yes, but it calls a third-party service you choose (your own API key or account) · partly (some features need the maker's servers) · no, it needs the maker's servers · not applicable - Do you need an account to use it? core
no account · optional account · account required - Who is it for? core
developers · non-developers · both - Is there public evidence it's built with AI coding agents? core
yes, the repo shows it (agent instruction files or agent co-authored commits) · yes, the maker says so · no public evidence - Which shared standard or format does it mainly build on? core
Model Context Protocol (MCP) · plain Markdown files (wikilinks, frontmatter) · a property-graph database (Neo4j, FalkorDB, Kuzu) · a SQL database (Postgres, SQLite) · none of these (its own format or interface) · not applicable - In the default setup, who mainly decides what gets remembered? value
the agent, by calling memory tools · the system, by extracting memories from conversations or data in the background · the app or user, by saving each memory explicitly - How can the person the memories are about see and edit them? value
directly, as readable files or in a UI they control · only through developer APIs or a developer dashboard · no documented way - By default, how do old or wrong memories mainly go away? value
only when someone deletes them (agent, app, or user) · the system updates, merges, or decays them automatically · they stay but are marked superseded (old facts are kept) - Where do memories live in the default setup the docs lead with? value
on the end user's own device · in the developer's own infrastructure · in the vendor's cloud - Does it work with models from different LLM vendors? value
yes, model-agnostic · no, tied to one model vendor - What is the main storage model for long-term memory? approach
vector store of extracted facts or chunks · knowledge graph (usually with vectors alongside) · plain files (Markdown, text, JSON) · structured records: profiles, tiers, or tables - What shape does it mainly take (the one its docs lead with)? approach
library or API primitive you build into your app · memory service (self-hosted server or hosted API) · agent runtime or harness with memory built in · end-user app, plugin, or assistant feature - Is there a first-party MCP server for it? approach
yes · no - How is memory separated between users or tenants? approach
built-in scoping by user, agent, session, or tenant IDs · one store per user or agent; the app partitions it - Does it publish scores on public memory benchmarks (LoCoMo, LongMemEval, BEAM, etc.)? approach
yes · no
Sources
Picked by hand (Phase 0); see the case study entry 'Whose communities?'.
43 sources across 22 projects, read on 2026-10-25. Each project page lists its own. Poll version: core-v4.